To increase sustainability in the field of machining with regard to energy efficiency improvement and resource conservation, accurate tool wear classification is of a paramount importance. In particular, optical wear monitoring approaches that use artificial intelligence and computer vision techniques have shown enormous potential in the recent past. A critical point to these approaches is the requirement of a lage image dataset of worn tools. The generation of such dataset is associated with a waste of resources, since recording of the data has to be done elaborately during ongoing production and classified by an expert. In this paper, the possibilities of generating synthetic training images with the help of generative adversarial networks (GANs) were investigated. A limited set of images was acquired for training GANs which then were used to increase the number of images for the training of our tool wear classificator. The approach enables a resource-efficient approach for the training process such that the accuracy and robustness of tool wear classification systems through the use of GANs can be ensured.